SolidWorks Flow Simulation

technical

PA P E R

ADVANCED BOUNDARY CARTESIAN MESHING TECHNOLOGY IN SOLIDWORKS FLOW SIMULATION Overview SolidWorks® Flow Simulation is an intelligent, easy-to-use computational fluid dynamics (CFD) program that facilitates the work of design engineers who use SolidWorks 3D CAD software for design creation. This paper details the approach and theory behind the Cartesian mesh generation used in SolidWorks Flow Simulation. Detailed explanation and examples will support how this choice of mesh generation algorithm enables SolidWorks Flow Simulation to offer a best-in-class CFD solution centered upon simplicity, speed, and robustness.

ABSTRACT For the numerical simulation of Navier-Stokes equations, the choice of the mesh type plays a significant role. Comparative calculations on different mesh types illustrates that the best simulation precision, characterized by minimum Local Truncation Error (LTE), is obtained on Cartesian meshes. For the boundary representation, the Immersed Boundary (IB) approach, which does not require a boundary-conforming mesh, is used. Use of Cartesian meshes together with the Immersed Boundary approach makes it possible to efficiently minimize approximation errors, build operators with good spectral properties so that robustness of method is guaranteed, speed up the process of grid generation, and make grid generation robust and flexible. Many other CFD methods require a mesh that fits the boundaries of the computational domain and often complex internal geometries. The body-fitted grid generation used is time-consuming, often requiring manual intervention to modify and clean-up of the CAD geometry as a prerequisite. Implementing the IB approach efficiently in SolidWorks Flow Simulation requires resolving a number of issues: approximation of the governing equations in cut-cells that contain the solid-fluid interface, capture of boundary layers effects irrespective of boundary layer thickness using a Two-Scale Wall Functions (2SWF) approach automatic mesh generation with automatic detection of initial mesh settings (octree-based mesh structure), and Solution Adaptive Refinement (SAR). Test cases given in this paper represent a small selection of our validation examples that illustrate the precision of the IB approach and the flexibility of SolidWorks Flow Simulation meshing technology in the wide range of industrial examples of geometry and physical formulations.

NOMENCLATURE

Subscript:

BREP Boundary Representation

q

Heating power

a

Ambient (pressure)

Cd

RRL

Result Resolution Level

c

Chamber (pressure)

Cmerge Unrefinement indicator value in Refinement Criterion

SAR

Solution-Adaptive Refinement

gap

Across the gap

Csplit Refinement indicator value in Refinement Criterion

t Time

i, i-1

Cell indices

δ Distance between the interpolated surface and the most distant point of the surface

SSF Small Solid Features Refinement Criterion

Drag coefficient

dgap

Minimal gap size (MGS)

dwall

Minimal wall thickness (MWT)

h

Mesh step size, i.e., cell size

l

Refinement level

L

Mesh refinement level

M

Mach number

LTE

Local Truncation Error

εsplit Refinement threshold value in Refinement Criterion

n

Mesh convergence coefficient

γ

Ratio of specific heats (Cp/Cv)

N

Number of cells

φ

Value of an angle

NURBS Non-Uniform Rational B-Spline p

Static pressure

δtolerance Threshold value of δ in Tolerance Refinement Criterion

Tol

Tolerance Refinement Criterion

x, y, z x, y, and z directions respectively

εmerge Unrefinement threshold value in Refinement Criterion

φcurvature Threshold value of φ in Curvature Refinement Criterion

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

2

1. Introduction SolidWorks Flow Simulation’s meshing technology has been in use for about 20 years, since the end of the 1980s. The task of developing a numerical flow simulation solution brought us to a decision point. The Navier-Stokes equations need to be discretized before they can be solved. However, before diving into the software development it was necessary to answer some important questions. Firstly, what mesh is the best for simulation precision? And secondly, what mesh is the best to use for practical applications? When the conservation laws for fluid flow are discretized on a spatial finite-volume mesh, the error in approximating these governing integral equations, together with the error in approximating the wall boundary conditions and flow boundaries, and therefore the simulation accuracy, is defined to a great degree by the choice of mesh geometry. The detailed reviews of basic types of mesh geometries are presented in several publications (see Weatherill & Hassan 1994, Filipiak 1996, and Parry & Tatchell 2008). These works show the wide use of body-fitted meshes for solving today’s industrial problems. As a rule, for more complicated geometries unstructured meshes are generated (Figure 1). In some cases the geometry allows the use of meshes with a regular structure. Such meshes are called structured meshes (Figure 2). In addition mixed-mesh topologies, structured in some subregions of computational domain and unstructured in other ones (Figure 3), are also used. Such meshes may be called partially structured or partially unstructured.

Figure 1 Unstructured body-fitted mesh

Figure 3 Combination of structured Cartesian mesh and hybrid unstructured body-fitted mesh near the wall

Figure 2 Structured body-fitted rectangular mesh

Figure 4 Structured Cartesian immersed-body mesh

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

3

The approach that uses the immersed-body mesh (Figure 4) is the alternative to the body-fitted approaches shown in Figures 1 to 3. Deliberately not conforming the mesh to the solid surface makes it possible to use a Cartesian mesh, which in general can be not body-fitted. Such mesh can be defined as the set of cuboidal rectilinear cells, which are adjacent to each other and the external boundary of the computational domain and orientated along the Cartesian coordinates. Cuboidal cells intersected by the surface (“cut-cells”) are treated according to the boundary conditions defined on the surface. The advantages of body-fitted meshes are as follows: • It is possible to place the mesh nodes optimally to resolve the geometrical features of the problem. This feature of body-fitted approach is exhibited most clearly in the case of unstructured meshes, which are the most “flexible” in terms of arbitrary placement of nodes during mesh generation. • It is possible to increase the mesh density towards the wall to ensure satisfactory resolution of the near-wall viscous layer. At the same time the body-fitted meshes are characterized by serious disadvantages caused by the following: • Approximation errors (Local Truncation Errors) caused by the spread of internodal distances (mesh irregularity) that is greatest for the unstructured meshes (see below in Table 1) • Loss of computational robustness, caused by the secondary fluxes at skewed faces, which increase the degree of schemes explicitness and, as a result, a decrease in the diagonal dominance of the difference operators • Serious difficulties in automating the process of mesh generation caused by the geometric complexity of the industrial problems; surface defects within the CAD representation; and the need to control the cell size and topology in the near-wall mesh to capture boundary layer development The examples of two structured quadrilateral and of one unstructured triangular meshes used to simulate the inviscid, isentropic, supersonic flow of ideal gas in a plane circular channel are presented below (see Aftosmis & Berger, 2002). Figure 5 displays the most typical variants of these meshes. Figure 5 (a) displays the “qualitative” unstructured near-equilateral triangular mesh, with angles that are not less than 29 degrees. Figure 5 (c) displays the structured quadrilateral curvilinear mesh, which is generated by the intersection of concentric circles with radial rays. Figure 5 (b) displays the structured mesh obtained from the curvilinear mesh Figure 5 (c) by the division of each quadrilateral cell into two triangular ones.

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

4

Figure 5 Body-fitted meshes: (a) Unstructured triangular mesh; (b) Structured triangular mesh; (c) Structured curvilinear quadrilateral mesh

Figure 6 displays the sequence of successively refined IB Cartesian meshes.

Figure 6 Sequence of Cartesian meshes with increased mesh density

(a) Ncells = 138

(b) Ncells = 507

(c) Ncells = 1928

(d) Ncells = 7549

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

5

For all mesh sequences presented above, Table 1 displays the values of Local Truncation Error (LTE) defined in the integral norm ||LTE||L , which are obtained by 1 applying the finite-difference operator that approximates the mass conservation law in the supersonic flow using the scheme of the formally second order of accuracy with respect to the accurate solution (Aftosmis & Berger, 2002). Here Cells is the number of mesh cells; n is the coefficient of mesh convergence. Table 1 Mesh

Results

n

Unstructured triangular

Structured triangular

Structured curvilinear

Cartesian [Aftosmis]

Cartesian [SolidWorks Flow Simulation]

Figure 5 (a)

Figure 5 (b)

Figure 5 (c)

Figure 6

Figure 6

Cells

||LTE||L

Cells

||LTE||L

Cells

||LTE||L

Cells

||LTE||L

Cells

||LTE||L

128

0.52552

144

0.37926

144

0.30998

138

0.03065

140

0.03014

505

0.22529

525

0.07571

525

0.09223

507

0.00930

516

0.00916

1918

0.11936

2001

0.01565

2001

0.02422

1928

0.00246

1944

0.00235

7490

0.05940

7809

0.00347

7809

0.00629

7549

0.00059

7526

0.00058

1

1.02

1

2.28

1

1.94

1

2.11

1

2.06

The presented results show clearly that the quality of approximation of the mass conservation law is significantly worse on non-Cartesian meshes: • The level of ||LTE||L is approximately 10 times higher on structured curvilinear 1 meshes than on a Cartesian mesh, and higher again on structured triangular meshes. The rate at which the errors reduce with increasing mesh count, characterized by the mesh convergence coefficient, n, is of the order of 2 for Cartesian and structured curvilinear and structured triangular meshes. • Compared against the Cartesian mesh, we see the level of ||LTE||L for the 1 unstructured triangular mesh is approximately 17 times higher on a coarse mesh, rising to 100 times higher on a fine mesh. The difference in the rate at which ||LTE||L reduces with reducing cell size is due to the difference in the mesh 1 convergence coefficient, n, between structured and unstructured meshes. As n is higher for the Cartesian mesh, ||LTE||L reduces faster as the Cartesian mesh is 1 refined (also see Figure 7). The difference between Cartesian and unstructured meshes will be even more pronounced in 3D, where tetrahedral cells are used for the unstructured mesh. To put this into context, based on this illustrative example, a typical industrial application in SolidWorks Flow Simulation using on the order of one million cells would require 100+ million cells in a tool that uses an unstructured tetrahedral mesh to achieve the same level of accuracy in representing the underlying conservation equations.

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

6

Figure 7 L1 norm of local truncation error in density for sequence of refined meshes (see Table 1, Aftosmis & Berger, 2002)

If the mesh contains skewed cells, in which the cell face is not normal to the line between the cell centers on each side of the cell face, then ||LTE||L is higher than 1 for a Cartesian mesh. This includes both structured and unstructured meshes. If the mesh topology is unstructured, with an irregular distribution of mesh nodes, then the mesh convergence coefficient, n, is reduced from around two to around one, so that the error reduces more slowly as the mesh size is increased. A Cartesian mesh is free from both of these disadvantages and can be regarded as the ideal mesh, producing the minimum ||LTE||L . 1

In conclusion, Cartesian meshes are characterized by their high approximation accuracy of the conservation laws. Let us examine the advantages and disadvantages of Cartesian meshes in general.

2. Advantages of the Cartesian Mesh 2.1 LTE Minimization Integral minimum of LTE follows from the fact that the structure of Cartesian mesh is maximally regular in the main area of the flow. Irregularity in the mesh is introduced at solid-fluid interfaces, which cause the regular cells to be split into a number of control volumes. The proportion of these cells reduces as the base mesh is refined. Another source of irregularity in the mesh is cells adjacent to a change in refinement level in the mesh. However, the proportion of these cells also reduces as the base mesh is refined. If N is the number of cells in a particular coordinate direction, then the model will typically contain of the order of N3 cells. The number of cut-cells is proportional to the surface area within the model, which is typically proportional to N2. The proportion of cut-cells is therefore proportional to N2/N3 = 1/N. A similar argument applies to the number of cells that are adjacent to a change in the mesh refinement level. Overall, irregularity is minimized the closer the mesh is to being fully Cartesian. It is then generally better from an accuracy perspective to use a finer base mesh with fewer refinement levels than vice versa; however, this will lead to a finer mesh overall.

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

7

2.2 Robustness of Differencing Scheme Robustness of the differencing scheme is ensured by the absence of secondary fluxes at skewed faces, since the Cartesian-based system ensures there are no skewed faces in the model. This in turn ensures maximum implicitness and diagonal dominance of difference operators in the matrices being solved. 2.3 Simplicity, Speed, and Robustness of Mesh Generation Algorithm Relative simplicity and a speed of automatic generation of Cartesian mesh are made possible by the application of the immersed-boundary mesh technique. CAD systems were originally developed for design process, not for numerical simulation needs. Consequently defects in CAD geometry (so-called “Dirty CAD”) are the common problem when using industrial CAD models for simulation (Figure 8). Many CAD systems use Boundary Representation (BREP) geometries to define 3D objects. A BREP surface is composed of a number of patches which can be quite complex, like in the case of a Non-Uniform Rational B-Spline (NURBS). This complexity is a source of nonwatertight joints resulting from imperfect edgeedge contact between adjacent surface patches. This may not be a problem when producing drawings or even rapid prototypes, but having a watertight (i.e., manifold) geometry is essential when creating a mesh for a flow simulation, especially in the case of body-fitted meshes which first require a surface mesh. Due to this, special efforts are necessary to repair or clean this kind of defect (Wang & Srinivasan, 2002 and Thompson et al., 1985) before meshing.

Figure 8 Typical defects view on CAD geometry

Body-fitted meshes are characterized by their high sensitivity to the quality of CAD geometry. Commonly, defects in the surface representation require user intervention to resolve the ambiguities and heal the defects of CAD geometry. Once this is done, a surface mesh is generated, usually beginning with nodes created on the solid surface. Then the surface triangulation is generated, most often by using a Delaunay algorithm, with point insertion as required. In addition, in some situations overrefinement of the surface can result in an excessive number of small triangles. This often happens in areas that are not important for flow simulation (for example, geometry features having a small radius, small spikes, joints and material interfaces, etc.). After that,

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

8

based on the surface triangulation, the volume mesh is generated. Often it is a mesh with tetrahedral elements that meet Delaunay criteria (see Delaunay 1934, Lawson 1977, Watson 1981, Baker 1989, and Weatherill & Hassan 1994). In the octree approach, when the initial mesh is generated, the geometry is always captured beginning with the large cells used for the background Cartesian mesh, and then with increasing precision as these are refined in accordance with criteria set. This approach makes it possible to stop the mesh refinement process once the mesh is locally fine enough to capture the physics of the problem. Unresolved geometry within a cut-cell can be completed by geometry interpolation inside the cell. Since the smallest cells needed to capture the physical detail present in the CAD model are relatively large compared to the size of any defects in the geometry, the sensitivity to CAD-geometry defects is much less. Cracks, overlaps, small components (for example, small spikes), and small defects in geometry definition can be ignored by using sufficiently large cells (Figure 9).

Figure 9 Defects are “healed“ by geometry interpolation inside the cell.

If there are narrow channels within the geometry that are important to the flow simulations, the required cell size can be achieved by mesh refinement up to the level corresponding to the prescribed characteristic width of the channel. Even in such cases it is typically the case that the dimensions of any CAD geometry defects are significantly less than the characteristic width of the channel. It is necessary to point out that the immersed body mesh approach could be implemented with tetrahedral and other types of elements (see Löhner et al., 2004), but in terms of approximation accuracy and ease of implementation, use of a Cartesian mesh is much better. 2.4 Simplicity of Navigation on Mesh In discussing the advantages of a Cartesian mesh we should also point out that the structured nature of the mesh results in much faster calculation of mesh-based information required by the solver, as well as speeding up the search for data associated with neighbor cells. Examples of such computations include the construction of cell geometry from the coordinates of its vertices, determining the solid volume and surface intersection with cell elements and faces, computation of spatial steps, and the location of cell centers. Searching for neighbor cell data is facilitated by the structured nature of the base mesh and the relationship between refined cells and their parents. Together these serve to speed up the computational efficiency. Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

9

2.5 Isotropic versus Anisotropic Refinement Mesh refinement can be implemented in a number of ways. Usually a tree structure is used, in which the larger cells are refined into smaller daughter cells. The most common approaches are binary tree refinement (Figure 10), where the parent cell is refined to two daughter cells, or octree refinement (Figure 11), where the parent cell is refined to eight equal daughter cells. There are other variants. On the one hand, the binary tree refinement is more universal and flexible, making it possible to generate the meshes better adapted to the solid geometry by using fewer cells.

Figure 10 Binary-tree refinement

Figure 11 Octree refinement

But practical attempts to implement such an approach (Berger & Aftosmis, 1998) show that the anisotropic refinement causes significantly irregular meshes, often with aspect ratios that decrease the solver accuracy. For 2D problems there is a benefit in terms of number of cells. But usually the typical number of cells for 2D problems is not too high; therefore, even if a 3D definition is used for solving a 2D problem, the computation time is not too high. In the case of arbitrary 3D geometry, such meshes are not sufficient to guarantee the proper resolution of the boundary layer due to the irregularity in the cell size. Additional issues with binary refinement include how to choose the refinement criterion and how to interface between regions with cells that have been refined in different orientations. Usually, the refinement criterion is calculated as the variation of the vector normal to the surface within the cell. If the variation in a specific direction is sufficiently high, the criterion is true and refinement takes place, splitting the parent cell into two new ones orthogonal to this direction. The refinement across the normal to the surface can cause extremely high step irregularity. Due to this, a more cautious refinement strategy would be needed, both initially and for solution adaptive refinement.

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

10

In terms of mesh topology there are some additional problems at the boundaries of “badly joined” refined cells. For example, if we refine two neighbor cells (Figure 12), in the way that one is split in the y direction and the other in the z direction, at the boundary we have face pairs which are not conformal.

Figure 12 Faces from the left and from the right side on x-face are different.

This topological situation requires additional efforts for interfacing, either by further refinement (thereby reducing the benefit of binary refinement on cell count), or by accepting a more complicated mesh structure that requires additional software logic (increasing the computation time per cell). If additional refinement is chosen, the mesh refinement process converges sufficiently quickly, but the enforced refinement of the neighbor cells causes the cell count to increase significantly, greatly reducing the benefit of this approach.

3. Cartesian Mesh Generation in SolidWorks Flow Simulation The technique used for mesh generation itself is strongly dependent on the properties and possibilities of the available solver and vice versa. The key challenges that arise if the Cartesian meshes are used are presented here along with their solution in SolidWorks Flow Simulation: 1) Treatment of sub grid-scale viscous near-wall layers, i.e., where the thickness of the boundary layer is less than or approximately equal to the characteristic size of mesh cell normal to the wall. This is solved through the use of a novel Two-Scale Wall Functions (2SWF) technique, where the subscale boundary layer is modeled using a special integral procedure. Otherwise, to solve the NavierStokes equations with two-equation k-ε turbulence would require a very fine computational mesh. Figure 13 displays the flow that includes the “thin,” “thick,” and “intermediate” layers, which are resolved universally by the 2SWF.

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

11

Figure 13 Velocity flow field with “ “thin,” “thick,” and “intermediate viscous boundary layer

2) Resolution of the wall, if the wall thickness is less than the characteristic size of mesh cell (i.e., “thin” walls) is achieved through the implementation of an original technique, in which the mesh cell can contain any number of “fluid” or “solid” control volumes, breaking the normal paradigm of a 1:1 correspondence between cells in the mesh and solver control volumes. 3) Figure 14 displays the velocity field around a solid, which includes segments of both “thick” and “thin” walls.

Figure 14 Velocity field near the solid, if there is a transition from “thick” to “thin” wall

In addition, there are several factors that result in irregular mesh step (and therefore a drop in the accuracy of the approximation): • Limited computational resources make it necessary to find the right balance between the total number of cells, the maximum permitted level of cell refinement (that allows to resolve the geometrical features), and the regularity of the mesh. In the case of arbitrary geometry the best choice is not evident. • An irregular base mesh. For maximum accuracy, the cell-cell contraction or expansion of the base mesh should not be very sharp. Usually the cell-cell size ratio should not exceed 1.2:1.

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

12

• A large difference in refinement level between adjacent cells. To avoid this issue, SolidWorks Flow Simulation restricts the difference in refinement level between a cell and its neighbor to be no more than one, forcing the refinement of neighbor cells where necessary. There are also some specific situations that arise which will influence the process of mesh generation, for example, in the case of narrow channels, sharp edges, solidsolid interfaces. We will discuss the most important aspects of this technique in the next section. 3.1 Initial Mesh Generation As was mentioned above, the Cartesian mesh based on octree technology is used in SolidWorks Flow Simulation. The mesh generation process begins with the definition of the rectangular computational domain. Three sets of planes orthogonal to the Cartesian coordinate system are then defined inside it. The intersection of these planes defines the set of rectangular cells (cuboids) that form the base mesh (Figure 15). The gaps between basic mesh planes can be varied. In addition, it is possible to specify the set of key planes, which are used to prescribe the position of some planes in the base mesh. Since it is necessary not to use a significantly irregular mesh, the restriction on the change in the spacing of the planes, h, is set to hi/hi-1 εsplit. If it is true, the cell has to be subdivided. The value of indicator Csplit is calculated individually for each cell according to an algorithm that controls that type of refinement. The value of εsplit is a constant defined automatically, but available to the user to change if desired. Setting refinement levels individually for each criterion makes it possible to influence the total number of cells that will be generated.

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

15

CAD, as a source of geometric information, provides a parametric definition of the solid surfaces. In SolidWorks Flow Simulation, the parametric definition is used to form the surface representation. Use of the CAD kernel’s parametric definition ensures that the approximated surface approaches the real surface as the mesh is refined. This approach has a further advantage over using a triangulated representation of the surface, being that it is possible to obtain the correct surface area and volume of the solid present in a cell, irrespective of how fine the mesh is. As such, this is also an advantage over body-fitted meshes, which require a fine surface mesh in order to correctly capture surface area and solid volume in regions of high geometric curvature. Refinement criteria can be categorized into the following groups: 1. Initial Geometry. Criteria that are used for initial capture of geometry: Tolerance, Curvature, and Small Solid Features. 2. Problem-specific. Criteria that deal with limitations and requirements caused by problem definition in local subareas: Cell Type, Solid Boundary, Selected Volume, Surface, and Edge refinements. 3. Narrow Channels. To increase the mesh quality in narrow channels, the special technique Narrow Channel Refinement has been developed. It ensures that an adequate number of cells are included to resolve the flow in the channel. Each refinement criterion is applied independently, although in some situations different criteria can result in the same refinement. If one or more refinement criteria are satisfied for a given cell, the cell is refined provided that the maximum number of refinement levels has not been exceeded. Surface triangulation, performed by the CAD system, is used to provide the information used in the refinement process. 3.2.1 Curvature Refinement Criterion The purpose of the curvature refinement criterion is to resolve geometrical details like corners and edges. According to this criterion, cells that contain geometry having a high degree of curvature within the cell are refined (Figure 19). The cell is tagged for refinement when the maximum variation of the angle between vectors normal to the surface triangles is greater than some critical value, and the limit level of curvature refinement has not been reached.

Figure 19 Curvature criterion: refinement is necessary, if normal vector variation within the cell φ>φcurvature (here: Csplit= φ and εsplit= φcurvature).

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

16

It is often convenient to use curvature refinement in cases where the mesh refinement needs fine-tuning. For example, if it is necessary to generate a more detailed mesh near a blunt angle which has not been refined sufficiently during the initial capture of geometry. 3.2.2 Tolerance Refinement Criterion The tolerance refinement criterion aims to improve the resolution of geometrical details like corners and edges within a single cell. This criterion identifies where the interpolated surface, formed from the intersection of the geometry with the faces of the cell, deviates too much from the actual geometry (Figure 20). This is done by testing whether the maximum distance between the interpolated surface and the most distant vertex of the surface triangles is greater than some critical value. Provided the limit on the number of levels of refinement for this criterion has not been reached, the cell is tagged for refinement.

Figure 20 Tolerance criterion: refinement is necessary if δ >tolerance (here: Csplit= δ and εsplit= δtolerance).

Note there is a difference between the curvature and tolerance refinement criteria. The curvature criterion will force refinement in regions of high curvature, associated with both blunt and sharp geometry. The tolerance refinement criterion allows the surface resolution to be improved if the interpolated surface does not go “deep enough” into the cell. Usually, the tolerance criterion value δ can be estimated approximately as 1/4min(dgap, dwall), where dgap and dwall are some values that characterize the minimal gap size and wall thickness available in the model. 3.2.3 Small Solid Features There are several geometrical situations when the criteria described above are insufficient to adequately resolve the geometry. One such example is when a relatively small solid, such as a cylinder or a sphere, is fully contained within one of the coarse mesh cells in the base mesh. Note that the tolerance criterion does not apply. Since the geometry does not intersect with the mesh, it is impossible to generate the interpolated surface to test against. The small solid features criterion is used for the initial surface capture before other refinement criteria are applied. The small solid features criterion is the same as the curvature criterion, but with threshold value of φcurvature=120° fixed in advance. The angle of 120° is chosen due to the fact that after triangulation of arbitrary 3D solids,

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

17

the maximum angle between the surface normals for the set of triangles always exceeds this value. In the extreme case of a sphere, where the triangulation creates a regular tetrahedron, the condition φcurvature=120° is true. As a result, the mesh sensitivity to small geometrical details is controlled just by this criterion alone. To illustrate how this criterion works it is helpful to consider an example. If, for the case presented in Figure 21, we deactivate all criteria, the small solid shown becomes invisible to the mesh. If we activate the tolerance criterion, it does not help the situation, since its operation requires an interpolated surface. If the small solid features criterion is activated and we impose the restrictions for refinement level LSSF = 2, the solid will be captured as shown in Figure 22. The size of the cell with refinement level 2 is of the same order as the solid characteristic dimension d which is enough for SolidWorks Flow Simulation to capture the geometry.

Figure 21 Body with the typical size d is invisible for the mesh level=0.

Figure 22 LevelSSF=2 cells are the same order of magnitude as the body size, d, small enough to capture the body.

In theory the same result could be obtained using the curvature criterion with the criterion value φcurvature set explicitly to 120°. However, using the small solid features criterion makes it possible to distinguish the primary process of coarse mesh capture of small solid features from the finer operations of the curvature and tolerance criteria, which are activated once the small solid geometry has already been captured. This allows these latter criteria to be adapted in accordance with the physical formulation of the specific problem. The operation of this criterion also helps in other situations (Figure 23) that are important for proper capture of the important details of the surface.

Figure 23 Small solid feature refinement works in the cells with φ≥120°.

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

18

The most important property of this criterion is that it generates refined cells only in the case where something exists inside the cell that is otherwise invisible to the mesh, or if there is some topologically important feature present on the surface. By using this criterion it is possible to generate a topologically correct mesh with the minimum number of cells. 3.2.4 Refinement in Local Regions The local mesh settings option is one more tool that helps to create an optimal mesh. The application of local mesh settings is especially beneficial if you are interested in resolving a particular region within a complex model. The local mesh settings can be applied to a component, whether it is an assembly, part, face, edge, or vertex. To apply the local mesh settings to some region, it is necessary to specify this region as a “virtual” solid part or subassembly (CAD systems allow the creation of “disabled” parts for this purpose) or to select an existing part or sub-assembly in the model. The local mesh settings are applied to cells intersected by the selected component, be it an assembly, part, face, edge, or just the single cell enclosing the selected vertex. 3.2.5 Narrow Channel Refinement The term “narrow channels” is conventionally used to describe a flow passage where the opposite walls are relatively close to each other (Figure 24).

Figure 24 Mesh refinement in narrow channel

The basic concept of the narrow channel refinement is to resolve the narrow channels with a sufficient number of cells to provide a reasonable level of solution accuracy. It is especially important to resolve the narrow channels in analysis of low Reynolds number flows and analyses involving long channels, where the boundary layer thickness becomes comparable with the size of the cut-cells. The mesh generator attempts to set the effective number of cells across the narrow channel as close as possible to the predefined target value of the approximate number of cells criterion (Ngap). Both the approximate number of cells criterion and the narrow channels refinement level (L) influence the mesh in narrow channels. The base mesh in the narrow channels will be split to have Ngap cells per channel width, provided that the refinement level of the cells generated in the narrow channel does not exceed L. The limit on the number of refinement levels is necessary to avoid the undesirable mesh splitting in very fine channels that may cause an unreasonably large number of cells.

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

19

This refinement criterion stands apart from the other criteria as it is not based on the local properties of cells intersected by solid geometry. Rather, it uses a raytracing method to detect the number of cells between solid surfaces that form a narrow channel and then tags these cells for refinement where necessary. Other control parameters used with this refinement criterion are the minimum and maximum heights of the narrow channels. These define the range of channels to which this criterion is applied. Outside this range the flow passage will not be considered as narrow and so will not be refined according to this criterion. 3.3 Automatic Mesh Settings As noted above, SolidWorks Flow Simulation’s octree mesh generator uses a system of refinement criteria, each with their own values and refinement levels. This is relatively straightforward compared to body-fitted mesh generators. However, in order to further simplify the process of defining the mesh, a method to automatically define the refinement criteria, values, and levels for a case Automatic Parameters Definition (APD) technology has been developed. There have been a number of improvements since the initial deployment of SolidWorks Flow Simulation’s APD technology from 1998 to 2000, most notably thin wall resolution. Consequently, the evolution of this approach requires a more lengthy description. The main idea of the APD technology is to define both the base mesh and refinement settings from the physical definition of the model’s boundary conditions and other elements, together with the most general information about the geometry. APD uses the following input parameters: 1. Problem type (Internal\External, Compressible\Incompressible, 3D\2D) 2. Minimum gap size (MGS) dgap 3. Minimum wall thickness (MWT) dwall 4. Level of initial mesh or Result Resolution Level (RRL) 5. Computational model bounding box 6. Fluid region bounding box 7. Symmetry settings applied to the computational domain boundaries 8. Middle size of the computational model 9. Wind direction for external flows APD output: 1. Base mesh a. Computational domain box BCD b. Key planes sets for x, y, z direction and cells ratios at these planes c. Number of cells for x, y, z direction Nx, Ny, Nz 2. Refinement settings a. Small solid feature refinement level LSSF b. Tolerance refinement level Ltol and tolerance dtol c. Narrow channels refinement level LNCR and the number of cells per channel width Ngap

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

20

The default value of MGS is the smallest size of all flow openings or faces where solution goals have been specified to control convergence of the solver. If set manually, MGS should be set to the smaller of the size of smallest flow passage to be resolved by the mesh, or the size of smallest geometric object in the flow that needs to be resolved. SolidWorks Flow Simulation is then able to calculate the refinement level needed to resolve this geometrical feature.

Figure 25(a) Minimum gap size

Figure 25(b) Minimum wall thickness Figure 25 Automatic initial mesh definition by using minimum gap size, minimum wall thickness, and level of initial mesh in order to calculate the needed levels of mesh refinement

Figure 25(c) Mesh setting window

Figure 25(d) Mesh view

The apd technology captures experience gained from manual mesh creation to provide values for all the necessary mesh refinement parameters for each result resolution level. In the case of the initial mesh, the result resolution level is the level of initial mesh shown in Figure 25(c) above. The result resolution level may change during the solution process due to the use of solution adaptive refinement in solidworks flow simulation. 3.4 Solution Adaptive Refinement The fundamental goal of Solution Adaptive Refinement (SAR) is to optimize the mesh distribution in order to minimize the spatial error arising from the discretization of the governing differential equations. In SolidWorks Flow Simulation this optimization process is implemented as part of the global iteration process. For a steady-state calculation, SAR cycles are implemented at a certain iteration frequency. Iterations are chosen on the basis of the estimated residual decrease at current iteration in comparison with residuals calculated after the previous SAR cycle. For transient cases, SAR cycles are implemented at certain moments in time.

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

21

The time approximation of conservation laws and difference scheme conservatism are ensured. In addition, history-dependent refinement is applied for such problems: once any cell has been subdivided, the generated subcells cannot be remerged. This approach makes it possible to refine the whole solution domain, where potentially any significant disturbances of the flow can appear. At each SAR cycle, the refinement level l is increased by one unless the certain predetermined maximum value is achieved. Cells that have been previously refined can be merged thus, reducing the refinement level by one. Two parameters, calculated from the local flow conditions, control the action of the SAR at a particular grid cell: a refinement indicator Csplit and an unrefinement indicator Cmerge. The Csplit and Cmerge parameters are calculated from the gradients present in the cell and in its immediate neighbors, so that Csplit takes into account not only the local gradient but also the number of neighbor cells and their size. To damp out numerical oscillations in the variable fields as the solution progresses, data used to calculate Csplit and Cmerge is collected over a number of iterations, and the collection does not start until the case has partially converged. If Csplit exceeds a certain predetermined value esplit then the LTE is expected to be high for the considered cell, and the cell is refined. For the set of eight daughter cells the unrefinement indicator Cmerge is introduced and its value calculated at the next SAR cycle. If Cmerge is then calculated to be less than a certain predetermined value emerge the LTE is expected to be low for the octet of cells and so they are merged. This approach was initially developed with first order indicators and applied to propulsion systems (see Gavriliouk et al. 1993, 1994a-1994c, Krulle et al. 1994 Barbashov et al. 1995, Hagemann et al. 1996, and Fey 2001), before being and extended to other applications. Just as with the initial mesh, to exclude the significant mesh irregularity in the regions of refinement the difference between the refinement levels of neighbor cells is limited to one. An example that demonstrates the efficiency of SAR in SolidWorks Flow Simulation is the stationary solution of the problem of supersonic ideal inviscid gas flow around the sphere with radius of one (M = 2 and γ = 1.4). Only one quarter of the sphere, cut by two planes of symmetry, is considered. Figure 26 displays the mesh and pressure distribution in the plane of symmetry. Refinement levels are l = 0, 1, 2, 3. It is seen that if l is increased, the resolution of the shock wave and the expansion fan are improved significantly and the lines of constant pressure correctly approximate circles. The information presented in Table 2 quantifies the efficiency of the technique.

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

22

Figure 26 Mesh and poressure distribution calculated for supersonic flow of ideal inviscid gas around a quarter of the sphere. A solution adaptive refinement was applied with the refinement levels of 0, 1, 2, 3. The images show the flow at symmetry plane.

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

23

Table 2 Refinement level, l

Total number Number of cells between of cells shock and spread point

Cd

Error %

0

2250

1.5

0.80374

1.86

1

7269

3

0.76207

-3.41

2

33316

6

0.78164

-0.93

3

143069

12

0.78672

-0.28

High-accuracy solution result (Lubimov & Rusanov, 1970): 0.7890 In Table 2 for each refinement level l the drag coefficient Cd for the front half (up to the angle of 90° relatively longitudinal axis) of the sphere is presented. Cd, calculated by SolidWorks Flow Simulation, is compared with an accurate solution obtained according to Lyubimov and Rusanov’s precise method [23]. It is seen that as l is increased from 0 to 3, the solution accuracy rises and by refinement level l =3, it is below one percent To obtain the same accuracy without SAR, i.e., using a uniform mesh, it would be necessary to use 1,152,000 cells—eight times higher than used above. These results show that the efficiency of SAR for external problems in SolidWorks Flow Simulation is very high.

4. INDUSTRIAL EXAMPLES 4.1 Pin-Fin Heat Sink Free convection cooling performance of a pin-fin heat sink has been studied experimentally (Yu & Joshi, 2002). The case of a 9x9 square pin-fin array presented in this paper has been investigated via SolidWorks Flow Simulation (Balakin & Churbanov, 2004) and compared with the above measurements. The model exactly reproduces the experimental conditions. It consists of two rectangular Plexiglas® enclosures, one inside another. The internal enclosure contains an aluminum pin-fin heat sink mounted on a heated component (Figure 27).

Figure 27 A close-up of the model: the internal enclosure with 9x9 pin-fin array over the heating component flush-mounted on the bottom

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

24

Two configurations are simulated; one with the heat sink mounted with the base horizontal, fins upward, and the other with the base vertical. Thermal conduction within the solid regions is modeled together with convection and radiative heat transfer. Figure 28 shows a comparison of experimental visualization (image on right) with predicted flow pattern for heat generation rate Q = 1 W (central image) on a vertical plane through the center of the heat sink for the model with the base mounted vertically. The mesh, shown in the image on the left, is obtained by SAR, and consists of about 145,000 cells. The images show a high level of agreement between simulation and experiment.





Figure 28 Mesh view (a), predicted flow trajectories colored by velocity magnitude (b) experimental visualization (c), Q = 1 W, vertical case, z = 0

The next set of images show the comparison of the predicted flow pattern with an experiment for the heat sink mounted horizontally. Figure 29 features a comparison of the experimental visualization (bottom image) with the predicted flow patterns for heat generation rate Q = 0.5 W (center image) on a vertical plane through the center of the heat sink. The mesh, obtained using SAR, is shown in the top image and consists of about 142,000 cells. A comparison of predicted and measured thermal resistance for considered cases shows that numerical results are within five percent of experimental data—an excellent correlation between computations and measurements. Note that the experimental results for the horizontal case are slightly nonsymmetrical. This is due to the difficulty in maintaining uniform boundary conditions on the outer Plexiglas enclosure.

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

25

Figure 29 Visualization from [24], Q = 0.5 W, horizontal case, z = 0 m

4.2 Arian Vulcain Engine Flow Simulation The nozzle jet of Arian rocket Vulcain is considered. The pressure ratio is pc / pa = 100 . Figure 30 shows the Mach number field (Fey, 2001) computed by SolidWorks Flow Simulation and an appropriately adapted Cartesian mesh, which consists of about 156,000 cells.

Figure 30 Mach number field in Vulcain nozzle and appropriate adapted Cartesian mesh computed by SolidWorks Flow Simulation technique (Fey, 2001). Pressure ratio is pc / pa = 100.

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

26

A very high level of accuracy is observed for SolidWorks Flow Simulation: • The high resolution of the shock generated by discontinuity of the second derivative of the profile of the nozzle wall on the line of interface of its mouth and extending part, which is a prerequisite for • Accurate computation of Mach disc size and position, being at the exit plane of the extending part, and • Accurate calculation of integral parameters: specific mass flow and specific impulse with relative error lying within 0.1 percent . 4.3 Flow around Spacecraft Progress-M Nose Numerical simulation of the supersonic flow around the nose part of the spacecraft Progress refers to the full model of the vehicle geometry, which includes superstructures (Figure 31), under external pressure of about 1 Pa. It is provided by SolidWorks Flow Simulation (Krylov et al., 2002) and the direct simulation with the Monte Carlo (DSMC) method (lvanov et al., 1998). Figure 32 displays the final SolidWorks Flow Simulation adapted mesh, which consists of about 1,000,000 cells, and Figure 33 shows the pressure fields computed by SolidWorks Flow Simulation (left) and the DSMC method (right).

Figure 31 Spacecraft geometry with superstructures. Pressure gauges P1 and P2 are located and seen on the right side of the surface.

Figure 32 The final SolidWorks Flow Simulation adapted mesh

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

27

Figure 33 Pressure fields computed by SolidWorks Flow Simulation (Krylov et al., 2002 [left]) and by direct simulation with the Monte Carlo (DSMC) method (Ivanov, et al. 1998 [right])

Comparison of these fields shown in Figure 33 shows they correspond very closely. Table 3 presents the relative pressures in gauges P1 and P2 at the moments t = 166.5s and t = 171.5s computed by SolidWorks Flow Simulation (Krylov et al., 2002) and by direct simulation with the Monte Carlo (DSMC) method (Ivanov et al., 1998). Table 3 Time

P1

P2

t = 166.5s

23.5 (DSMC)

42.5 (DSMC)

23.6 (SolidWorks Flow Simulation)

47.2 (SolidWorks Flow Simulation)

11.5 (DSMC)

25.5 (DSMC)

12.7 (SolidWorks Flow Simulation)

28.5 (SolidWorks Flow Simulation)

t = 171.5s

It can be seen that the SolidWorks Flow Simulation data and DSMC data are in very good agreement: their difference is less than 10 percent.

5. Conclusions The general purpose CAD-embedded CFD solver in the SolidWorks Flow Simulation software package from SolidWorks uses an octree-based mesh technology, combined with a unique immersed boundary approach for wall friction and heat transfer. Totally free from mesh distortion, this Cartesian-based approach has been shown to deliver the lowest error when the Navier-Stokes equations are discretized onto the mesh. SolidWorks Flow Simulation’s octree refinement allows far more rapid changes in mesh density than can be achieved with tetrahedral meshes, reducing computation time. Octree is particularly suited to solution adaptive

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

28

refinement, for example, to capture shocks in high-speed flows. Excellent agreement has been shown for several industrial examples that have included buoyancy-driven flow, conjugate heat transfer (convection, conduction, and radiation), and hypersonic flow with shock capture.

REFERENCES Aftosmis, M.J. and Berger, M.J., 2002. Multilevel Error Estimation and Adaptive h-Refinement for Cartesian Meshes with Embedded Boundaries. AIAA 2002-0863. 40th AIAA Aerospace Sciences Meeting and Exhibit. January 14-17, Reno, NV Baker, T.J., 1989. Automatic Mesh Generation for Complex Three-Dimensional Regions Using a Constrained Delaunay Triangulation, Engineering with Computers, Vol. 5, pp.161-175 Balakin, V., Churbanov, A., Gavriliouk, V., Makarov, M., and Pavlov, A., 2004. Verification and Validation of EFD.Lab Code for Predicting Heat and Fluid Flow, Proceedings of CHT-04 ICHMT International Symposium on Advances in Computational Heat Transfer, April 19-24, Norway, CHT-04-179 Barbashov, E., Gavriliouk, V., Guriev, V., Isaev, V., Losenkov, A., Odintsov, E.V., Sergienko, A.A., Sobachkin, A.A., 1995. Mathematical Modeling of Propulsion Systems, IAF 95-S.1.05, 46th International Astronautical Congress, October 2-6, Oslo, Norway Berger, M.J., Aftosmis, M.J., 1998. Aspects (and Aspect Ratios) of Cartesian Mesh Methods, Proceedings of the 16th International Conference on Numerical Methods in Fluid Dynamics, to appear in “Lecture Notes in Physics,” Springer−Verlag, Heidelberg, Germany, July 6−10, Arcachon, France Delaunay, B.N., 1934. Sur la Sphere Vide, Izvestia Akademia Nauk SSSR, VII Seria, Otdelenie Matematicheskii i Estestvennyka Nauk, Vol. 7, pp. 793-800 Filipiak, M., 1996. Mesh Generation, Edinburgh Parallel Computing Centre, University of Edinburgh, Version 1.0, November 1996 Fre, M. 2001. Behandlung von Strömungsproblemen in Raketendüsen bei Überexpansion, Institut für Aerodynamik und Gasdynamik, Universität Stuttgart, Germany, 2001 Gavriliouk, V.N., Denisov, O.P., Nakonechny, V.P., Odintsov, E.V., Sergienko, A.A., Sobachkin, A.A., 1993. Numerical Simulation of Working Processes in Rocket Engine Combustion Chamber, 44th Congress of the International Astronautical Federation, IAF-93-S.2.463, October 16-22, Graz, Austria Gavriliouk, V.N., Krulle, G., Schley, C.-A., and others, 1994b. Numerical simulation of combustion process in rocket combustion chambers with coaxial injection, AIAA Paper 94-3329.

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

29

Gavriliouk, V.N., Krulle, G., Odintsov, E., Schley, C.-A., Sergienko, A., Sobatchkine, A., 1994c. Numerical Simulation of Combustion Process in Rocket Combustion Chambers with Coaxial Injectors, AIAA 94-3329, 30th AIAA/ASME/SAE/ASEE Joint Propulsion Conference, June 27-29, Indianapolis, IN Gavriliouk, V.N., Lipatnikov, A.V., Kozlyaev, A.N., Odintsov, E.V., Sergienko, A.A., Sobachkin, A.A., 1994a. Computation Modeling of the Combustion Problems with the use of “AeroShape-3D,” Numerical Technique-ISTS 04-d-27 Hagemann, G., Schley, C.-A., Odintsov, E., Sobatchkine, A., 1996. Nozzle flowfield analysis with particular regard to 3D-plug cluster configurations, 32nd AIAA/ASME/ SAE/ASEE Joint Propulsion Conference and Exhibition, July 1-3, Lake Buena Vista, FL Ivanov M.S., Markelov G.N., Gimelshein S.F., 1998. Statistical simulation of reactive rarefied flows: numerical approach and applications, AIAA Paper 98-2669 Krulle, G., Gavriliouk, V., Schley, C.-A., Sobachkin, A., 1994. Numerical simulation of Aerodynamic Processes and its Applications in Rocket Engine Problems, 45th Congress of the International Astronautical Federation, IAF 94-S2.414, October 9-14, Jerusalem, Israel Krylov, A.N., Kotov, V.M., Tokarev, V.A., Shcherbakov, N.A., Khokhlov, A.V., Ivanov, M.S., Vaschenkov, P.V., Kashkovsky, A.V., and Markelov, G.N., 2002. Numerical Modeling and Experimental Data Analysis of the Flow Near Spacecraft Progress-M Nose after the Head Fairings Jettisoning, ESA SP Vol. 487 pp. 307-314 Lawson, C.L., 1977. Software for C1 Surface Interpolation, Mathematical Software III, pp. 161-194 Löhner, R., Cebral, J., Castro, M., Baum, J.D., Luo, H., Mestreau, E., and Soto, O., 2004. Adaptive Embedded Unstructured Grid Methods, Mecánica Computacional, Vol. XXIII, pp. 29-42, G. Buscaglia, E. Dari, O. Zamonsky (Eds.), Bariloche, Argentina, November Lubimov, A.N., Rusanov, V.V., 1970. Techenija gasa okolo tupykh tel. Gas flows around blunt bodies, Vol. I, Nauka Press, Moscow (in Russian) Mentor Graphics Corp., 2011. Enhanced Turbulence Modelling in SolidWorks Flow Simulation Parry, J. and Tatchell, D., 2008. Flomerics’ EFD Meshing Technology: A White Paper Thompson, J.F, Warsi, Z.U.A, Martin, C.W., 1985. Numerical Grid Generation: Foundations and Applications. North-Holland: New York, NY Watson, D.F., 1981. Computing the Delaunay Tesselation with Application to Voronoi Polytopes. The Computer Journal, Vol. 24(2) pp. 167-172

Advanced Boundary Cartesian Meshing Technology in SolidWorks Flow Simulation

30

Weatherill, N.P. and Hassan, O., 1994. Efficient Three-Dimensional Delaunay Triangulation with Automatic Point Creation and Imposed Boundary Constraints, International Journal for Numerical Methods in Engineering, Vol. 37, pp. 2005-2039 Wang Z.J. and Srinivasan, K., 2002. An adaptive Cartesian grid generation method for “Dirty” geometry, International Journal for Numerical Methods In Fluids Yu, E. and Joshi, Y., 2002. Heat Transfer Enhancement from Enclosed Discrete Components Using Pin-Fin Heat Sinks, International Journal of Heat and Mass, Vol. 45, No. 25, pp. 4957-4966 To learn more about how SolidWorks Flow Simulation solutions can help you develop better products, please contact your local SolidWorks reseller or visit www.solidworks.com.

Dassault Systèmes SolidWorks Corporation 175 Wyman Street Waltham, MA 02451 USA Phone: 1 800 693 9000 Outside the US: +1 781 810 5011 Email: [email protected] www.solidworks.com

SolidWorks is a registered trademark of Dassault Systèmes SolidWorks Corporation in the US and other countries. Other brand and product names are trademarks or registered trademarks of their respective owners. ©2013 Dassault Systèmes. All rights reserved. MKBCMeshwpENG0213